Method and apparatus for providing a trained machine learning model for detecting a hands-on / hands-off state on a vehicle's steering wheel
The teacher-student training approach for steering torque detection simplifies the model complexity and resource needs, enhancing the efficiency of hands-on/hands-off state recognition in vehicle systems.
Patent Information
- Application Number
- DE102024200002
- Authority / Receiving Office
- DE · DE
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2024-01-02
- Publication Date
- 2024-12-05
- Estimated Expiration
- 2044-01-02
AI Technical Summary
Existing steering torque-based detection methods for determining a hands-on/hands-off state on a vehicle's steering wheel face challenges due to noisy torque measurements influenced by various factors, leading to large and resource-intensive machine learning models.
A method and device utilizing teacher-student training to develop a simplified student model for hands-on/hands-off detection, using general and specialized teacher models to adapt to specific contexts, reducing model complexity and resource requirements while maintaining functional quality.
The simplified student model achieves improved functional quality with reduced memory and computing resources, enabling efficient hands-on/hands-off state recognition for vehicle assistance systems.
Smart Images

Figure 00000000_0000_ABST
Abstract
Description
[0001] The invention relates to a method and a device for providing a trained machine learning model for recognizing a hands-on / hands-off state on a vehicle's steering wheel. Furthermore, the invention relates to a method and a device for recognizing a hands-on / hands-off state on a vehicle's steering wheel.
[0002] Vehicles use sensors, such as a capacitive steering wheel, to monitor driver activity. Such a steering wheel detects whether the driver is touching ("hands-on") or not touching ("hands-off") the steering wheel using a capacitive sensor. This information is then transmitted to relevant functions, such as longitudinal and / or lateral guidance assistance systems. The presence of hands on the steering wheel indicates driver activity and attentiveness. For example, the system might prompt the driver to place their hands on the steering wheel if it detects that their hands have been off the wheel for a predetermined period during lateral guidance.
[0003] To save on the additional costs of a capacitive sensor in the steering wheel, it is known to monitor driver activity using machine learning models, in particular artificial neural networks, based on torque (hand torque) detected at the steering wheel. Such a method is known, for example, from DE 10 2019 211 016 A1.
[0004] DE 10 2021 129 058 A1 generally refers to machine learning and in particular to training a compact model that performs the same function as a larger model.
[0005] A major challenge of steering torque-based detection systems is identifying the driver-induced steering torque within the measured (noisy) steering torque. Many factors can contribute to noisy steering torque, particularly the sensor's position (it typically forms part of the steering gear or power steering system, creating a torsional vibration system via the steering column's elasticity in conjunction with the steering wheel, whose inherent dynamics complicate the precise measurement of driver-induced torque); the degree of friction within the steering system; feedback from road irregularities; the weight of the steering wheel / steering system; and steering wheel vibration caused by an assistance function (e.g., haptic feedback when leaving the lane).
[0006] Furthermore, the characteristics (used for hands-off detection) of the measured steering torque can change due to external influences, e.g., temperature, vehicle load, the presence of a trailer, tire type and / or tire condition, steering system changes over its lifespan, road gradient / slope / incline, etc.
[0007] Due to the number of influencing factors, the machine learning model, especially the neural network, needs to be equipped with enough degrees of freedom, i.e., weights, so that the model can represent all these influences. However, this leads to a large model and thus a very high demand for storage and computing resources. In particular, the relationship between the number of weights and the accuracy of the representation of the desired objective function is not linear, but increases almost exponentially.
[0008] The use of teacher-student training is known, for example, from DE 11 2020 001 663 T5. Furthermore, the use of teacher-student training is known from G. Hinton et al., Distilling the Knowledge in a Neural Network, arXiv:1503.02531v1 [stat.ML], March 9, 2015.
[0009] The invention is based on the objective of improving a method and a device for providing a trained machine learning model for recognizing a hands-on / hands-off state on a vehicle's steering wheel. Furthermore, the invention is based on the objective of improving a method and a device for recognizing a hands-on / hands-off state on a vehicle's steering wheel.
[0010] The problem is solved according to the invention by a method with the features of claim 1 and a device with the features of claim 9. Furthermore, the problem is solved by a method with the features of claim 8 and a device with the features of claim 10. Advantageous embodiments of the invention are set forth in the dependent claims.
[0011] In particular, a first aspect provides a method for providing a trained machine learning model for recognizing a hands-on / hands-off state on a vehicle's steering wheel, comprising training at least one general teacher model using a training dataset, training at least one context-specific teacher model, whereby training data of the context is selected from the training dataset and / or the training dataset is enriched with training data of the context, training a student model using the general teacher model and / or the at least one specialized teacher model by means of teacher-student training, wherein the selection of which teacher model(s) are used to train the student model is based on a context corresponding to the respective training date, and providing the trained student model.
[0012] Furthermore, in a second aspect, a device for providing a trained machine learning model for recognizing a hands-on / hands-off state on a vehicle's steering wheel is provided, comprising a data processing device, wherein the data processing device is configured to train at least one general teacher model using a training data set, to train at least one context-specific teacher model, and for this purpose to select training data of the context from the training data set and / or to enrich the training data set with training data of the context, to train a student model using the general teacher model and / or the at least one specialized teacher model by means of teacher-student training, and a selection of which teacher model(s) are used to train the student model.to make decisions based on a context corresponding to the respective training date and to provide the trained student model.
[0013] Furthermore, in particular in a third aspect, a method for detecting a hands-on / hands-off state on a vehicle's steering wheel is provided, comprising detecting at least one steering parameter on the vehicle's steering wheel, using a student model trained according to the embodiments of the first and / or second aspect to detect the hands-on / hands-off state, wherein the detected at least one steering parameter is supplied to an input of the trained student model for this purpose, and providing the hands-on / hands-off state estimated by the trained student model.
[0014] Finally, in a fourth aspect, a device for detecting a hands-on / hands-off state on a steering wheel of a vehicle is provided, comprising at least one sensor for detecting at least one steering parameter on a steering wheel, a control unit, wherein the control unit is configured to receive the detected at least one steering parameter, to provide and apply a student model trained according to the embodiments of the first and / or second aspect in order to detect the hands-on / hands-off state, and to supply the detected at least one steering parameter to an input of the trained student model for this purpose, and to provide the hands-on / hands-off state estimated by the trained student model.
[0015] The method and device described in the first and second aspects allow for the provision of a model with improved memory and computing resource consumption. For this purpose, a student model is provided, which, in terms of its structural design, is simpler and less complex than the teacher models used. For example, the student model can have fewer neurons or fewer internal layers, and thus fewer parameters (especially weights), than the respective teacher models. To prevent a reduction in functional accuracy, it is planned that, in addition to a general teacher model, at least one context-specific teacher model will be trained.In teacher-student training, the training of a student model is carried out using the general teacher model and / or at least one specialized teacher model. The selection of which teacher model(s) to use for training the student model is based on a context corresponding to the respective training date. The student model trained in this way is then made available, specifically output. This trained student model is used, in particular, in a vehicle to detect a hands-on / hands-off state. Based on this, for example, an assistance system can be operated.
[0016] A particular advantage of the method and the device is that an improved student model with improved functional quality can be provided. This allows for the provision of a machine learning model with reduced storage and computing resource requirements, especially when functional quality is improved.
[0017] The models (the teacher models and the student model) are specifically trained to estimate a hands-on / hands-off state based on at least one control variable.
[0018] The at least one general teacher model, the at least one specialized teacher model, and the student model are designed as artificial neural networks, specifically deep neural networks, which in particular comprise several inner layers. The models are specifically artificial recurrent neural networks that process the input data X at each time t. tprocess and at least a hands-off probability y t Output in [0,1]: y t = p(x t |x 0:t-1Preferably, the neural networks output a probability for a hands-on state and a hands-off state at each time t. In particular, the neural networks possess a so-called memory h, in which information from previous time steps is stored and which can be used for output at the current time step. The output is further processed, for example, by filtering, before the receiving functions (e.g., a lateral guidance assistant) process it. Specifically, it can be provided that, based on a comparison of the hands-off probability with a predefined threshold, a binary hands-off signal is generated (with the two states "hands-off detected" and "hands-off not detected"). A similar process can be implemented for the hands-on state, or a single signal with the two signal states "hands-on" or "hands-off" can be provided.
[0019] During training (i.e., during the respective training phase), the models are trained using training data comprising pairs in which steering parameter data, in particular torque data, are paired with a hands-on / hands-off state as the baseline. The data of the at least one steering parameter, in particular the torque data, are specifically time series of the at least one steering parameter measured at the steering wheel, specifically time series of torques measured at the steering wheel. The training data are obtained, in particular, by means of test drives and / or in simulators. In principle, the provision of training data can be carried out, in particular, according to the procedure described in DE 10 2019 211 016 A1. The training is carried out in a manner known per se, in particular by means of supervised learning. The acquisition and / or collection of the training data set can be part of the first and second aspects.The hands-on / hands-off state is detected in particular by means of additional sensors (e.g. by means of a capacitive steering wheel).
[0020] The Loss function calculates and optimizes the error per unit of weight during backpropagation training. In particular, binary cross-entropy can be used when the hands-on / hands-off state is detected. L=−y*log(Y)+(1−y)*log(1−Y), where Y is the ground truth and y is the output of the model.
[0021] When training the student model, a modified loss function is used: L=λ*student−loss+(1−λ)*distillation−loss, where the student loss is the binary cross-entropy between the output of the student model and the baseline truth, and the distillation loss is the binary cross-entropy between the output of the student model and the teacher model. The weighting factor λ indicates the strength of the teacher model's influence compared to the influence of the original training data (i.e., the baseline truth). The weighting factor λ is determined, for example, using empirical trials.
[0022] In teacher-student learning, the student model is trained using at least one general teacher model and / or at least one specialized teacher model. Both the teacher and student models are fed the same training data as input. In addition to a basic truth value, an output from the teacher model, or a combination of teacher models, is used to adjust parameters (especially weights) of the student model and thus train it. The adjustment of these two parameters is achieved using the previously mentioned adjusted loss function. When combining teacher models, the outputs of the teacher models are weighted accordingly.
[0023] The method for providing a trained machine learning model for recognizing a hands-on / hands-off state on a vehicle's steering wheel can be implemented particularly as a computer-implemented method.
[0024] A steering parameter is, in particular, a parameter that represents and / or describes the current state of the steering wheel. A steering parameter is, in particular, a torque, which is detected, especially by means of a torque sensor on the steering wheel. However, a steering parameter can also be any other parameter detected directly or indirectly at the steering wheel. For example, it might be possible to detect the current in an electric motor at the steering wheel and use it as a steering parameter. Detecting the hands-on / hands-off state can be based solely on the steering parameter detected at the steering wheel, in particular a detected torque. However, it is also possible for the machine learning model to be provided with other (steering) parameters detected at the steering wheel (e.g., steering wheel angle and / or steering wheel angular velocity, etc.).The trained machine learning model can then recognize the hands-on / hands-off state, taking this additional parameter(s) into account. Furthermore, parameters not measured at the steering wheel can also be considered, such as vehicle speed, lateral acceleration, yaw rate, wheel ticks, damper information, and / or other vehicle dynamics parameters, etc. However, no capacitive sensor is specifically planned for the steering wheel.
[0025] A hands-off state is, in particular, a state in which the driver does not touch the steering wheel. Specifically, none of the driver's fingers are in contact with the steering wheel. Detecting the hands-off state can, in particular, include providing a hands-off state signal. This signal may include, for example, a hands-off probability or coded signals for the states "hands-off detected" and "hands-off not detected." A hands-on state is, in particular, a state in which the driver touches the steering wheel. Specifically, a hands-on / hands-off state is provided, for example, as a hands-on / hands-off state signal with, in particular, two signal states (e.g., "hands-on detected" or "hands-off detected").
[0026] A context refers to, or encompasses, in particular, the characteristics of a situation in which the detection of the hands-off state is to take place and / or in which one or more values of at least one steering parameter have been recorded. Examples of characteristics that can define a context include: an outside temperature (e.g., specialized instructor models may be provided for extremely hot or extremely cold temperatures), an inside temperature, steering wheel vibration, the vehicle's load and / or weight, the presence of a trailer, the presence of snow chains, cobblestones, potholes, maximum steering interventions (e.g., steering wheel vibration), speed bumps, driver characteristics (e.g., identity, gender, age, weight), etc. The current context is recognized and / or determined, in particular, based on recorded sensor data.For example, it may be possible to query such sensor data via a vehicle's CAN bus and / or obtain it from the vehicle's sensors and / or control system when using the trained student model.
[0027] A subdivision of contexts and the corresponding provision of specialized teacher models can, for example, be based on expert knowledge. In particular, the need for specialized teacher models for different contexts or aspects of a context can be determined based on expert knowledge. However, automated classification methods can also be used for this purpose.
[0028] Parts of the devices, in particular the data processing units, can be designed individually or collectively as a combination of hardware and software, for example as program code that runs on a microcontroller or microprocessor. However, it is also possible for parts to be designed individually or collectively as an application-specific integrated circuit (ASIC) and / or a field-programmable gate array (FPGA).
[0029] In one embodiment, the trained student model is loaded into the memory of at least one vehicle control unit (ECU) after training. This allows the trained student model to subsequently be used in the vehicle to provide hands-on / hands-off detection. Loading the model into memory includes, in particular, loading a structural description and parameters (weights, hyperparameters, etc.) of the student model into the memory. Specifically, an assistance system of the vehicle can be operated based on a detected hands-on / hands-off state.
[0030] In one embodiment, the at least one specialized teacher model is initialized at the beginning of the training process using parameters of the trained general teacher model. This reduces the training time of the at least one specialized teacher model, as it is already pre-trained due to the initialization.
[0031] In one embodiment, selection is based on metadata associated with the respective training data. This allows the context in which the training data is located to be directly determined based on the metadata linked to that data. Such metadata can, for example, directly describe the context, such as the circumstances under which the training data, and in particular the at least one control variable, was recorded. Examples of context properties have already been described above.
[0032] In one embodiment, the selection process is carried out at least partially using a decision tree. This allows one or more teacher models to be selected directly, depending on contextual factors. It is also possible to use only the general teacher model during teacher-student training, for example, if no specialized teacher model is available for a given training date.
[0033] In one embodiment, the selection is performed at least partially by means of a further trained machine learning model, wherein the further trained machine learning model is trained to select the most suitable teacher model or models for teacher-student training based on training data supplied as input data. This allows suitable teacher models or combinations of teacher models to be found for all training data.
[0034] In a further developmental embodiment, it is provided that the further machine learning model is trained using the trained teacher models, wherein the teacher model or those teacher models which exhibit the highest accuracy in estimating the hands-on / hands-off state are assigned as the basic truth to each training date.
[0035] Further features regarding the design of the devices are derived from the description of embodiments of the respective method. The advantages of each device are the same as those of the embodiments of the respective method.
[0036] Furthermore, a steering system is also created, comprising a device for detecting a hands-on / hands-off state on a steering wheel of a vehicle according to one of the described embodiments.
[0037] Furthermore, a vehicle is also created, comprising a steering system according to one of the described embodiments and / or a device for detecting a hands-on / hands-off state on a steering wheel of a vehicle according to one of the described embodiments.
[0038] The invention is explained in more detail below with reference to preferred embodiments and the figures. These show: Fig. 1 a schematic representation to illustrate embodiments of the device for providing a trained machine learning model for detecting a hands-on / hands-off state on a vehicle's steering wheel; Fig. 2 a schematic flowchart to illustrate embodiments of the method for providing a trained machine learning model for detecting a hands-on / hands-off state on a vehicle's steering wheel; Fig. 3 A schematic representation to illustrate embodiments of the method for providing a trained machine learning model for detecting a hands-on / hands-off state on a vehicle's steering wheel.
[0039] The Fig. Figure 1 shows a schematic representation to illustrate embodiments of the device 1 for providing a trained machine learning model for recognizing a hands-on / hands-off state 40 on a steering wheel of a vehicle 50. The device 1 is configured to perform the method described in this disclosure.
[0040] The device 1 comprises a data processing unit 2. The data processing unit 2 comprises at least one computing unit 2-1 and at least one memory 2-2. The computing unit 2-1 can perform arithmetic operations on data stored in the memory 2-2 to carry out the procedure.
[0041] The data processing unit 2 is configured to train at least one general teacher model 20 using a training dataset 10, to train at least one teacher model 21-x specialized with respect to a context 11-x, and for this purpose to select training data 10-x of the context 11-x from the training dataset 10 and / or to enrich the training dataset 10 with training data 10-x of the context 11-x, and to train a student model 22 using the general teacher model 20 and / or the at least one specialized teacher model 21-x by means of teacher-student training. Furthermore, the data processing unit 2 is configured to select which teacher model(s) 20, 21-x are used to train the student model 22, based on a context 11-x corresponding to the respective training data 10-x.The student model 22 trained in this way is provided by the data processing facility 2.
[0042] It can be provided that, after training, the trained student model 22 is loaded into a memory 53 by at least one control unit 52 of a vehicle 50 as part of the provisioning process. The control unit 52 is part of a device 51 for detecting a hands-on / hands-off state 40 at a steering wheel (not shown) of a vehicle 50. The device 51 further comprises at least one sensor 54 for detecting at least one steering parameter 55 at the steering wheel. The at least one steering parameter 55 includes, in particular, a torque. The control unit 52 is configured to receive the detected at least one steering parameter 55, to provide and apply the trained student model 22 in order to detect the hands-on / hands-off state 40, and to supply the detected at least one steering parameter 55 to an input of the trained student model 22 for this purpose, and to output the hands-on / hands-off state 40 estimated by the trained student model 22.The hands-on / hands-off state 40 can then be fed to another control unit 56 of the vehicle 50 for further processing. In particular, it may be possible to operate an assistance system of the vehicle 50 based on the hands-on / hands-off state 40. This could, for example, be a lateral guidance assistant.
[0043] The Fig. Figure 2 shows a schematic flowchart illustrating embodiments of the method for providing a trained machine learning model for detecting a hands-on / hands-off state on a vehicle's steering wheel. The method can, for example, be implemented using a device such as the one described above. Fig. 1 was described.
[0044] In measure 100, at least one general teacher model is trained using a training dataset. Specifically, a (single) general teacher model is trained. The training is carried out in a manner known per se, in particular by means of supervised learning.
[0045] In measure 101, training data is selected from the training dataset according to its context. Furthermore, training data can be enriched according to its context.
[0046] In measure 102, at least one teacher model specialized with regard to a specific context is trained. In particular, a large number of specialized teacher models can be trained, each specialized for a particular context.
[0047] In measure 103, a student model is trained using the general teacher model and / or at least one specialized teacher model through teacher-student training. Measure 103 comprises measures 103a to 103d.
[0048] In step 103a, a training date is selected from the training dataset. This can be done randomly and / or in groups. In step 103b, a suitable teacher model (general or specialized) or a group of teacher models (general and / or specialized) is selected for the chosen training date. This is done taking into account the context of the training date. In particular, properties of the context are considered. For example, a context can have the characteristics already described in the general description. In this way, the most suitable teacher model or combination of teacher models for each training date can be used. In step 103c, the student model is trained in a teacher-student training process in a manner known per se.Step 103d checks whether the training of the student model is complete, i.e., whether a predetermined number of training runs and / or training data has been reached. Furthermore, the process may include assessing the quality of the student model and, depending on the results, either continuing or ending the training. If the check in Step 103d indicates that the training should not yet be stopped, the process returns to Step 103a and continues with additional training data. Conversely, if the check in Step 103d indicates that the student model training is complete, the process proceeds to Step 104.
[0049] Measure 104 involves providing the trained student model. Specifically, it is intended that, as part of the provisioning process, the trained student model will be loaded into the memory of at least one vehicle control unit. Subsequently, a vehicle assistance system can be operated based on the outputs of the trained student model.
[0050] It may be provided that at least one specialized teacher model 21-x ( Fig. 1) is initialized at the beginning of training using parameters of the trained general teacher model 20.
[0051] It can be provided that the selection is based on metadata 12-x linked to the respective training date 10. Using the metadata 12-x, properties and characteristics of the context 11-x can be directly described (e.g., trailers present / not present; weather conditions, etc.). Based on the metadata 12-x, corresponding metadata 12-x can be searched for, which are assigned to the teacher models 20, 21-x and which allow a selection of the teacher model 20, 21-x or the teacher models 20, 21-x.
[0052] It may be possible to perform the selection at least partially using a decision tree. This allows a suitable teacher model 20, 21-x or a suitable combination of teacher models 20, 21-x to be selected directly for a training date 10-x, based on the context 11-x, which is described, for example, in the form of metadata 12-x.
[0053] It may be provided that the selection is carried out, at least partially, by means of a trained additional machine learning model 13, wherein the trained additional machine learning model 13 is trained to select, based on a training data 10-x supplied as input data, the teacher model 20, 21-x or the teacher models 20, 21-x most suitable for teacher-student training. The trained additional machine learning model 13 is stored in memory 2-2 (in particular in the form of a structure description and associated parameters) and is provided by the data processing unit 2.
[0054] It may be further provided that the further machine learning model 13 is trained using the trained teacher models 20, 21-x, wherein a respective training date 10-x is assigned the teacher model 20, 21-x as the basic truth or those teacher models 20, 21-x are assigned as the basic truth which(s) exhibit(s) the highest accuracy in estimating the hands-on / hands-off state 40.
[0055] The Fig. Figure 3 shows a schematic representation illustrating embodiments of the method for providing a trained machine learning model for recognizing a hands-on / hands-off state on a vehicle's steering wheel. The schematic sequence of the teacher-student training is shown. A trained general teacher model 20 and several trained specialized teacher models 21-x are available to train the student model 22.
[0056] During teacher-student training, models 20, 21-x, and 22 are supplied with training data 10-x. This means that at each training step, models 20, 21-x, and 22 are provided with a training data set 10-x, which is the same for all models 20, 21-x, and 22. A training data set 10-x, for example, includes the input data x1 to x. n , which in particular include measured values of at least one control variable at various, successive time steps. A training date 10-x corresponds to a context 11-x in which this training date 10-x was recorded or in which this training date 10-x was generated (e.g., by means of a simulation).
[0057] For example, the specialized teacher model 21-1 may be a teacher model specialized for context 11-x, in which the vehicle has a trailer attached to a trailer hitch. Furthermore, the specialized teacher model 21-2 may be a teacher model specialized for context 11-x, in which there is a vibration at the steering wheel, for example, applied by an actuator to provide haptic feedback to the driver.
[0058] During teacher-student training, a suitable teacher model 20, 21-x, or a suitable combination of teacher models 20, 21-x is selected, taking into account the context 11-x corresponding to the respective training date 10-x. For example, if context 11-x, in which a trailer is attached, corresponds to training date 10-x, the specialized teacher model 21-1 is selected. In the example of context 11-x, in which there is a vibration in the steering wheel, the specialized teacher model 21-2 is selected accordingly. If the context 11-x is not quite so clear, the general teacher model 20 or a combination of teacher models 20, 21-x can be selected. For example, if a trailer is attached to the vehicle and there is a vibration in the steering wheel, both specialized teacher models 21-1, 21-2 can be used, with the outputs 14-x being weighted accordingly.
[0059] The student model 22 is trained using output 14-x of the selected teacher model 20, 21-x, or a combination of selected teacher models 20, 21-x. For this purpose, the student model 22 is fed the same training data 10-x, and the student model 22 generates an output 15 based on this. This output 15-x is then compared with output 14-x (by way of knowledge distillation 17) and a basic truth 16-x, which is available for the training data 10-x. Based on a loss function, as described in the general description section, the parameters (weights) of the student model 22 are adjusted.
[0060] It is therefore specifically intended that the teacher models 20, 21-x are selected for each training step taking into account the respective context 11-x of the training date 10-x present in this training step. Reference symbol list 1 Device 2 Data processing equipment 2-1 Computing Equipment 2-2 storage 10 Training data set 10-x Training date 11-x context 12-x Metadata 13 further machine learning models 14-x edition (teacher model / s) 15-x edition (student model) 16-x Fundamental Truth 17 Distillation of Knowledge 20 general teacher model 21-x specialized teacher model 22 Student model 40 Hands-On / Hands-Off State 50 vehicles 51 Device 52 Control unit 53 storage 54 Sensor (steering parameter) 55 Steering size 56 additional control unit (assistance system) 100-104 measures of the student model procedure (22) to provide an estimated hands-on / hands-off state (40).
Claims
[1] A method for providing a trained machine learning model for detecting a hands-on / hands-off state (40) on a steering wheel of a vehicle (50), comprising: Training at least one general teacher model (20) using a training data set (10), Training at least one teacher model (21-x) specialized with respect to a context (11-x), wherein for this purpose training data (10-x) of the context (11-x) are selected from the training data set (10) and / or the training data set (10) is enriched with training data (10-x) of the context (11), Training a student model (22) by means of the general teacher model (20) and / or the at least one specialized teacher model (21-x) by means of a teacher-student training, wherein a selection of which of the teacher models (20, 21-x) are used to train the student model (22) is made based on a context (11-x) corresponding to the respective training date (10-x), Providing the trained student model (22). [2] Method according to claim 1, characterized by that the trained student model (22) is loaded into a memory (53) of at least one control unit (52) of a vehicle (50) after the training as part of the provision. [3] Method according to claim 1 or 2, characterized by that the at least one specialized teacher model (21-x) is initialized at the beginning of training using parameters of the trained general teacher model (20). [4] Method according to one of the preceding claims, characterized by that the selection is based on metadata (12-x) associated with the respective training date (10-x). [5] Method according to one of the preceding claims, characterized by that the selection is made at least partly by means of a decision tree. [6] Method according to one of the preceding claims, characterized by that the selection is carried out at least partially by means of a trained further machine learning model (13), wherein the trained further machine learning model (13) is trained to select the teacher model (20,21-x) or the teacher models (20,21-x) most suitable for the teacher-student training on the basis of a training date (10-x) supplied as input data. [7] Method according to claim 6, characterized bythat the further machine learning model (13) is trained using the trained teacher models (20,21-x), wherein the teacher model (20,21-x) which has / have the highest quality in estimating the hands-on / hands-off state (40) is / are assigned as ground truth to a respective training date (10). [8] A method for detecting a hands-on / hands-off state (40) on a steering wheel of a vehicle (50), comprising: Detecting at least one steering variable (55) on the steering wheel of the vehicle (50), using a student model (22) trained according to one of claims 1 to 7 to detect the hands-on / hands-off state (40), wherein the detected at least one steering variable (55) is supplied to an input of the trained student model (22) for this purpose, and Providing the hands-on / hands-off state (40) estimated by the trained student model (22). [9] Device (1) for providing a trained machine learning model for detecting a hands-on / hands-off state (40) on a steering wheel of a vehicle (50), comprising: a data processing device (2), wherein the data processing device (2) is designed to to train at least one general teacher model (20) using a training data set (10), to train at least one teacher model (21-x) specialized with respect to a context (11-x), and for this purpose to select training data (10-x) of the context (11) from the training data set (10) and / or to enrich the training data set (10) with training data (10-x) of the context (11-x), to train a student model (22) by means of the general teacher model (20) and / or the at least one specialized teacher model (21-x) by means of a teacher-student training, and a selection of which of the teacher models (20, 21-x) are used to train the student model (22), based on a context (11-x) corresponding to the respective training date (10-x), and to provide the trained student model (22). [10] Device (51) for detecting a hands-on / hands-off state (40) on a steering wheel of a vehicle (50), comprising: at least one sensor (54) for detecting at least one steering variable (55) on a steering wheel, a control unit (52), wherein the control unit (52) is configured to receive the detected at least one steering variable (55), to provide and apply a student model (22) trained according to one of claims 1 to 7 in order to recognize the hands-on / hands-off state (40), and to supply the detected at least one steering variable (55) to an input of the trained student model (22) for this purpose, and to provide the hands-on / hands-off state (40) estimated by the trained student model (22).
Citation Information
Patent Citations
Detection of hands-off situations through machine learning
DE102019211016A1
KNOWLEDGE DISTILLATION BY MEANS OF DEEP CLUSTERING
DE102021129058A1
Autonomous vehicle system
DE112020001663T5
Cited By
Method and device for generating training data for training a machine learning model for hands-off detection
DE102025113011A1
Method and apparatus for providing a robustized machine learning model for detecting a hands-off state on a vehicle's steering wheel
DE102025113890A1
Method and device for providing a trained machine learning model for detecting a hands-on / hands-off state of a steering wheel of a vehicle
WO2025146338A1
